Force on ForcePrime Rogue Inc.
A grognard exercise, not a scientific model. What that means

Roadmap

Every addition to the game, newest first. Each one is a git tag: what was added, what it found, what it cited, and the commit that did it.

Tags and commit hashes refer to the project's git repository; check out any tag to get the model exactly as it was. Line counts leave out result tables and generated data.

Findings are quoted as they were written at each version. Later versions change the numbers, and sometimes overturn the finding. The current state of each module is on the modules page.

AheadWhat comes next, and why

ItemWhyRaised in
Training that carries into the resistanceThe model uses training only in the opening fight. A volunteer with one week counts the same as one with six months once the war becomes an occupation.0.1.3.7
Casualty tolerance that depends on mobilizationCanadian deaths push a mobilized Canada toward capitulation exactly as they push an unmobilized one. That makes extra fighting look self-defeating.0.1.3.7
Provincial governments as actorsWhether a province's leaders accommodate an occupier is the largest lever the provincial work found, and it is currently a switch set by hand.0.1.3.5
Resistance that grows from a quiet startResearch on occupations of rich urban societies says violence usually starts low and grows. The model starts resistance at its level.0.1.3.5
Partial annexation as a war aimThe US chooses all of Canada or Alberta. It cannot yet choose to hold a set of provinces.0.1.3.4
A calibrated capitulation hazardThe chance Canada's government gives way drives the result in long wars and rests on no case base. It needs occupied and coerced states from 1938 onward.0.1
Out-of-sample occupation backtestsThe occupier death rate was tuned to Iraq 2003 to 2011. It needs Afghanistan, Vietnam, Soviet Afghanistan and Chechnya.0.1
Who gets blamedSeveral results (blackouts, a shot-down patrol, polar bears) turn on which government the public blames. That is set by hand in each module.0.1.2.5
Other dyadsThe engine does not know it is about Canada. Candidates are Denmark and Greenland, Venezuela, Taiwan and the Baltics.0.1

Done29 versions

0.1.3.8
18 Nov 2026
tag v0.1.3.8, commit 0be98fd
8 files, +810 −2 lines
“tools/publish_frontend.sh: copy the built site into the front-end repository”
Tooling

The public site, and what each module is worth now

This site, generated from the repository. Every module measured again, on against off, in three scenarios. No change to the model.

Switch: none. The model is unchanged from v0.1.3.7; results are identical.

What it found, as written at this version

  1. A public site now presents the whole project: every version with its tag and commit, every parameter with its source line, the method and its limits. It is generated from the repository, so it cannot drift from the code.
  2. 173 of the model's 224 parameters are the modeller's judgement. The site says so on every page and marks each one. This is a grognard exercise, not a scientific model.
  3. Allies and mobilization carry the shooting scenarios; potash and oil carry the one without shooting. The table below is the current worth of every module.

Runs: 3,000 paired runs per variant, three scenarios, every switch removed in turn from the all-on baseline (about 250,000 runs).

Full results document

0.1.3.7
18 Nov 2026
tag v0.1.3.7, commit 3b5a178
12 files, +1,301 −1 lines
“v0.1.3.7: results doc test count”
Analysis

What the 400,000 does, and why

The mobilization plan taken apart. How much of it is needed, each of its three effects switched off in turn, arming, training, the US response, and allies.

Switch: canada_mobilization.mode (unchanged).

What it found, as written at this version

  1. All of the plan's value in a full annexation comes from its effect on Canada's will to continue. With that effect switched off, the 400,000 make Canada's odds worse than today (−2.0 points). With only that effect, the gain is +5.5.
  2. That effect is one hand-set parameter. It cuts the chance Canada capitulates by 10 to 40% and is complete at half fill. If it grows in proportion to the size of the plan instead, the full plan is worth +12.1 points, not +4.3.
  3. The first half of the plan does the work. Gains are +0.9, +2.2, +5.1, +4.6 and +4.3 points at 10%, 25%, 50%, 75% and 100% of the way from today to the full plan.

Runs: 3,000 paired runs per variant, every other module on, v0.1.3.5 defaults. 222,000 new runs; 285,000 with v0.1.3.6. 32 tests pass.

Full results document   Video

0.1.3.6
17 Nov 2026
tag v0.1.3.6, commit 7a6f189
5 files, +149 −10 lines
“v0.1.3.6: mobilization levels (today's reserves, plan as sampled, full 400,000) and comparison”
Analysis

Today's reserves against the 400,000 plan

Three levels of Canadian mobilization compared. Today's reserves, the plan as the model samples it, and the full 400,000.

Switch: canada_mobilization.mode (unchanged).

What it found, as written at this version

  1. Against today's reserves, the full 400,000 plan raises the chance the US withdraws by 4 points in a full annexation (67% to 71%) and by 9 points in an Alberta detachment (54% to 63%).
  2. Nearly all of that gain arrives by the time the plan is about half filled. Going from the half-filled plan the model normally assumes to the full 400,000 adds nothing measurable in a full annexation and under 2 points in Alberta.
  3. The full plan costs more lives on both sides. US deaths rise 42% over today's case and Canadian deaths 39%.

Runs: 3,000 paired runs per variant, every other module on, v0.1.3.5 defaults. 31 tests pass; harness checks pass.

Sources cited at this version (4)
  • Supplementary Reserve 4,384: May 2025 mobilization directive, as reported; the Ottawa Citizen still called it "the current 4,384" on 16 November 2026.
  • DND's Reserve Force components page says "about 6,700", undated.
  • No source was found for 4,135.
  • The same Ottawa Citizen article gives 25,780 part-time Primary Reservists; the model holds 23,561 (May 2025). Not changed in this version.

Full results document

0.1.3.5
17 Nov 2026
tag v0.1.3.5, commit 382117c
6 files, +297 −12 lines
“v0.1.3.5: province pacification calibrated from research (delay, hard provinces, relapse, accommodation excurs…”
Module

Province pacification set from research

How fast a province goes quiet, set from a commissioned research report. A delay of 12 to 18 months, hard provinces that hold out, relapse as a monthly chance.

Switch: spatial.enabled.

What it found, as written at this version

  1. The open question from v0.1.3.4 is closed, and the answer is "slow". That version showed the result swinging by up to 10 points on how fast one province goes quiet. The research puts the speed near the slow end, and rules out the fast calibration. Across the whole range the research allows, US withdrawal in a full annexation moves by about half a point either way.
  2. Full annexation barely changes: 71.4% US withdrawal (71.7% in v0.1.3.4). Alberta detachment (61.6%) and coercion (34.4%) do not change at all.
  3. The reason is timing. No province can go quiet in its first 12 to 18 months, and the average war in the model is decided in 23. Ontario goes quiet at any point in only 5% of runs.

Runs: 3,000 paired runs per variant, every other module on. 30 tests pass; all harness checks pass (26, 26 and 22), including the check that the uniform split still equals the single-area model.

Sources are listed parameter by parameter under Provinces.

Full results document

0.1.3.4
17 Nov 2026
tag v0.1.3.4, commit aa377e2
10 files, +414 −17 lines
“v0.1.3.4: results doc wording”
Module

Where the US puts its troops, and provinces pacified one at a time

Three rules for allocating occupation troops (by need, population first, resources first) and pacification province by province.

Switch: spatial.enabled.

What it found, as written at this version

  1. The provincial split now moves outcomes, and the direction depends on one number nobody has: how fast a single province goes quiet. With the slower of two defensible calibrations, pacifying province by province helps Canada by 1 to 3 points. With the faster one it helps the US by 5 to 10.
  2. Concentrating troops cuts US deaths by roughly a quarter under either calibration. Whether it also wins depends on the same number.
  3. Putting resource provinces first cuts the US gasoline shock by a third (30 cents a gallon to 20). Putting population centres first makes it half as bad again (45 cents), because the oil is left unheld.

Runs: 3,000 paired runs per variant, every other module on. 29 tests pass; all harness checks pass, including the check that the uniform split still equals the single-area model.

Sources are listed parameter by parameter under Provinces.

Full results document

0.1.3.3
13 Nov 2026
tag v0.1.3.3, commit 2d08125
9 files, +185 −14 lines
“v0.1.3.3: resources by province (oil, potash, electricity shares; held share = provincial share x control, rep…”
Module

Resources by province

Oil, potash and electricity attached to the provinces that produce them. What the US holds of each is each province's share times US control there.

Switch: spatial.enabled (resources are part of the default; spatial.resources: false gives the earlier yes/no flags).

What it found, as written at this version

  1. The only visible change is in full annexation: the old flag said the US could hold 92% of the oil, and the provincial shares say all of it is in the theatre. The Alberta scenario's hand-set 84% matches Alberta's actual 83.8%.
  2. This version is plumbing. Its use is that the model now knows what each province is worth, so that a partial occupation gets the right resources automatically and the US has something to choose between in the next version.

Runs: 3,000 paired runs per variant, every other module on. 28 tests pass; all harness checks pass.

Sources cited at this version (5)
  • Canada Energy Regulator, "Market Snapshot: Canada sets new record in crude oil production in 2025" (https://cer-rec.gc.ca/en/data-analysis/energy-markets/market-snapshots/2026/market-snapshot-canada-sets-new-record-in-crude-oil-production-in-2025.html).
  • Statistics Canada, The Daily, 22 October 2025, electricity exports by province in 2024 (https://www150.statcan.gc.ca/n1/daily-quotidien/251022/dq251022c-eng.htm).
  • Statistics Canada, "Electricity generation: combustibles and renewables sharply increase amid another dry year" (2025 exports from Manitoba and Quebec) (https://www.statcan.gc.ca/o1/en/plus/9117-electricity-generation-combustibles-and-renewables-sharply-increase-amid-another-dry-year).
  • RBC, "Power play: assessing Canada's electricity advantage in US trade talks" (New Brunswick's 2023 share; Quebec's historical share) (https://www.rbc.com/en/thought-leadership/the-trade-hub/power-play-assessing-canadas-electricity-advantage-in-u-s-trade-talks/).
  • Export values: v0.1.2.3 to v0.1.2.5 data files (Canada Energy Regulator 2025 trade snapshots; potash from Natural Resources Canada).

Full results document

0.1.3.2.1
13 Nov 2026
tag v0.1.3.2.1, commit 04831b6
12 files, +42 −42 lines
“v0.1.3.2.1: Alberta scenario armed propensity set from the research (0.27/0.49/0.76, was 0.15/0.30/0.50), all …”
Data correction

Alberta baseline corrected

The Alberta scenario's armed propensity reset from the research (0.27, 0.49, 0.76; was 0.15, 0.30, 0.50). Written up with v0.1.3.3.

Switch: spatial.enabled (resources are part of the default; spatial.resources: false gives the earlier yes/no flags).

What it found, as written at this version

  1. Kevin approved setting the Alberta scenario's armed propensity from the research: 0.27 / 0.49 / 0.76 in place of 0.15 / 0.30 / 0.50, in all 11 Alberta scenario files. New Alberta baseline with every module on:
  2. Every earlier Alberta result is about 2 points lower than it would be on the new baseline.

Runs: 3,000 paired runs per variant, every other module on. 28 tests pass; all harness checks pass.

Sources cited at this version (5)
  • Canada Energy Regulator, "Market Snapshot: Canada sets new record in crude oil production in 2025" (https://cer-rec.gc.ca/en/data-analysis/energy-markets/market-snapshots/2026/market-snapshot-canada-sets-new-record-in-crude-oil-production-in-2025.html).
  • Statistics Canada, The Daily, 22 October 2025, electricity exports by province in 2024 (https://www150.statcan.gc.ca/n1/daily-quotidien/251022/dq251022c-eng.htm).
  • Statistics Canada, "Electricity generation: combustibles and renewables sharply increase amid another dry year" (2025 exports from Manitoba and Quebec) (https://www.statcan.gc.ca/o1/en/plus/9117-electricity-generation-combustibles-and-renewables-sharply-increase-amid-another-dry-year).
  • RBC, "Power play: assessing Canada's electricity advantage in US trade talks" (New Brunswick's 2023 share; Quebec's historical share) (https://www.rbc.com/en/thought-leadership/the-trade-hub/power-play-assessing-canadas-electricity-advantage-in-u-s-trade-talks/).
  • Export values: v0.1.2.3 to v0.1.2.5 data files (Canada Energy Regulator 2025 trade snapshots; potash from Natural Resources Canada).

Full results document

0.1.3.2
13 Nov 2026
tag v0.1.3.2, commit 4bce5a1
12 files, +335 −13 lines
“v0.1.3.2: political geography by province (opposition and armed-propensity multipliers from the 13 Nov 2026 re…”
Module

Political geography by province

Provincial differences in opposition and armed propensity, with correlated blocs and wartime compression. Inputs from a commissioned research report.

Switch: spatial.enabled (political geography is now part of the default; spatial.politics: false gives v0.1.3.1).

What it found, as written at this version

  1. Provinces now start with different levels of resistance, and it changes nothing about who wins. The chance the US quits is identical with and without political geography in every one of 3,000 runs, in both occupation scenarios, and stays identical under every excursion tried (Quebec at either end of its range, Alberta or the Prairies accommodating, Atlantic Canada and the North at the top of theirs).
  2. That is a property of how the model is built, not evidence that provincial politics is irrelevant. The reason is below, and it says what has to come next.
  3. The one result that does move an outcome is about the Alberta scenario's own input: the research does not support the low armed propensity that scenario assumes. Using the research value raises the chance the US quits Alberta by 2.4 points (1.7 to 3.1) and US deaths by 36%.

Runs: 3,000 paired runs per variant, every other module on. 27 tests pass; all harness checks pass.

Sources cited at this version (8)
  • Daoust, "The 51st State? Exploring Canadian Attitudes Towards US Annexation and Their Determinants", Canadian Journal of Political Science, 16 June 2026 (https://www.cambridge.org/core/journals/canadian-journal-of-political-science-revue-canadienne-de-science-politique/article/51st-state-exploring-canadian-attitudes-towards-us-annexation-and-their-determinants/FDC9E14505B5FBAF6A7C01E6DDDA9302).
  • Leger, Alberta Pulse Check, June 2026; Leger national polls, December 2024 and March 2025 (via CBC, https://www.cbc.ca/news/canada/edmonton/canada-u-s-51st-state-annexation-1.7490806).
  • Angus Reid Institute, January 2025 (https://angusreid.org/canada-51st-state-trump/) and July 2025 (https://angusreid.org/canada-volunteer-military-service/).
  • Ipsos for Global News, January 2026 (https://www.ipsos.com/en-ca/half-canadians-would-join-preparation-program-if-invasion-occurs).
  • Environics for the Cascade Institute, 2025 and 2026 (https://cascadeinstitute.org/willing-to-fight/).
  • RCMP, 2025 Commissioner of Firearms Report (https://rcmp.ca/en/corporate-information/publications-and-manuals/2025-commissioner-firearms-report).
  • Elections Alberta official referendum results, as reported by CBC (https://www.cbc.ca/news/canada/calgary/alberta-referendum-official-results-9.7401791).
  • Research Co., December 2024 (https://researchco.ca/2024/12/20/unity-canada-3/).

Full results document

0.1.3.1
13 Nov 2026
tag v0.1.3.1, commit 1b04e2b
14 files, +354 −47 lines
“v0.1.3.1: population and geography by province (real 2026 populations, rural dispersion, presence floor per la…”
Module

Population and geography by province

Real 2026 populations, rural dispersion, a presence floor by land area, regional weather and local shocks.

Switch: spatial.enabled (now the geography version by default; spatial.uniform: true keeps the v0.1.3.0 skeleton for the aggregation check).

What it found, as written at this version

  1. Geography moves where the troops are, not who wins. With all five attributes on, the chance the US quits changes by +0.1 points in full annexation (−0.1 to +0.3) and 0.0 in the Alberta scenario (−0.2 to +0.2).
  2. I said this would be the first version in which provinces diverge. They diverge in troops, a lot. They barely diverge in control (one point), and the reason is structural, explained below.

Runs: 3,000 paired runs per variant, every other module on. 26 tests pass; all harness checks pass, including the aggregation check.

Sources cited at this version (3)
  • Statistics Canada, population estimates for 1 July 2026 (https://www150.statcan.gc.ca/n1/daily-quotidien/260923/dq260923b-eng.htm).
  • Statistics Canada, Census in Brief, "Population growth in Canada's rural areas, 2016 to 2021" (https://www12.statcan.gc.ca/census-recensement/2021/as-sa/98-200-x/2021002/98-200-x2021002-eng.cfm): rural population and rural share by province.
  • Statistics Canada, 2021 Census, Table 98-10-0001-01: land area. The 13 figures used sum to the published national land area of 8,788,702.80 km²; Ontario, Quebec and Nunavut were checked individually, the other ten from my recall of the same table, validated only by that sum.

Full results document

0.1.3.0
13 Nov 2026
tag v0.1.3.0, commit e8ec40f
13 files, +308 −15 lines
“v0.1.3.0: provinces as separate areas of occupation (skeleton): 11 units, StatCan July 2026 populations, propo…”
Module

Provinces as separate areas: the skeleton

Ten provinces and the North as separate areas of occupation, with no differences between them yet. It exists to pass one test, that the split model reproduces the single-area model.

Switch: spatial.enabled (occupation scenarios only; it has no path in economic coercion).

What it found, as written at this version

  1. The differences are floating-point rounding. The golden results did not change. Running 11 areas instead of one adds about 3% to run time.
  2. The harness now has a permanent aggregation check (check command): split versus single area, same outcome and end month in every run, and every output within one part in a billion.

Runs: 3,000 paired runs per scenario, every module on. 25 tests pass; all harness checks pass.

Sources cited at this version (3)
  • Statistics Canada, Table 17-10-0009-01 and The Daily, 23 September 2026 (https://www150.statcan.gc.ca/n1/daily-quotidien/260923/dq260923b-eng.htm).
  • Manitoba Bureau of Statistics, Quarterly Population Statistics Report, 23 September 2026 (https://www.gov.mb.ca/mbs/publications/mbs304_pop_2026_m07.pdf): provincial figures.
  • Territorial figures as reported from the same Statistics Canada release (https://www.gofarglobal.com/news/canada-population-2026); they close the sum to the national total exactly.

Full results document

0.1.2.14.1
13 Nov 2026
tag v0.1.2.14.1, commit 9b2c085
5 files, +199 −8 lines
“v0.1.2.14.1: film a run as an MP4 (web/tools/film.py); clearer diary wording for allied reversals and second-w…”
Tooling

A run as a video

A script that films a run of the replay page, and clearer war-diary wording for allied reversals and second-war theatres.

No results document. No change to the model.

0.1.2.14
12 Nov 2026
tag v0.1.2.14, commit e6b53c9
12 files, +1,397 −5 lines
“v0.1.2.14: per-month run traces, web export, static replay front end (web/)”
Tooling

Run traces and the replay page

A month-by-month record of every run, and a static page for watching single runs against all 2,000.

What it found, as written at this version

  1. No change to the model. Results are identical to v0.1.2.13.

Full results document

0.1.2.13
12 Nov 2026
tag v0.1.2.13, commit 35e0d48
“v0.1.2.12 + v0.1.2.13: allied air cover without ground troops; a second war against a US ally (two switches, t…”
Module

A second war against a US ally

Another state attacks a US ally while the US is committed in Canada. Shares a commit with v0.1.2.12.

Switch: second_war.enabled.

What it found, as written at this version

  1. A second war is good for Canada in almost every variant. The US cannot hold Canada and fight elsewhere without the pressure to leave Canada rising sharply. The effect at the modelled hazard is small (2 to 3 points) only because the war is unlikely in any given run.
  2. A Pacific war helps Canada most, because it costs the US economy most (5% of GDP a year) and takes the most air power. In the coercion scenario, where money is the whole contest, it raises the chance the US gives up by 31 points.
  3. Europe is the one case that can hurt Canada. If Europe is attacked and the US stays out, Canada loses European materiel and gains nothing: −0.5 points in full annexation. If the US defends Europe, Canada still gains (+8.7), but less than from other theatres, because Europe stops supplying Canada and half the time lifts its sanctions on the US. Canada's allies trade Canada for their own defence.

Runs: 3,000 paired runs per variant, every other module on except allied air. 23 tests pass; all harness checks pass.

Sources are listed parameter by parameter under A second war.

Full results document

0.1.2.12
12 Nov 2026
tag v0.1.2.12, commit 35e0d48
12 files, +615 −19 lines
“v0.1.2.12 + v0.1.2.13: allied air cover without ground troops; a second war against a US ally (two switches, t…”
Module

Allied air cover without ground troops

Allies offer a combat air patrol over Canada. The US respects it or engages it.

Switch: allied_air.enabled.

What it found, as written at this version

  1. The patrol matters as a tripwire, not as air power. A respected patrol at realistic coverage moves the outcome by about half a point. A patrol the US shoots down moves it by 4 to 12 points, because allied dead cost Washington more at home and abroad than the patrol ever cost it in the air.
  2. So the US's best move is to leave the patrol alone, and the allies' offer has most of its value only if Washington is expected to shoot. The model's 50% chance of that is the single most important number here and the least supported.
  3. Warning is what makes air cover real. Coverage triples with three months' notice. A respected patrol then adds 2 to 5 points, mostly by keeping Canada's government in place and materiel flowing.

Runs: 3,000 paired runs per variant, every other module on except the second war. 23 tests pass; all harness checks pass.

Sources are listed parameter by parameter under Allied air cover.

Full results document

0.1.2.11
12 Nov 2026
tag v0.1.2.11, commit 02380ea
13 files, +332 −23 lines
“v0.1.2.11: NATO split, member by member (correlated tier draws from the deep-research report; continuous allie…”
Module

NATO split, member by member

Each ally drawn separately with correlated blocs. Tripwire troops, warning time and US counter-moves. Inputs from a commissioned research report.

Switch: nato.model: members (the harness baselines now use it; coalition keeps the v0.1.2.8 single draw). nato.scenario_class: S1 economic coercion, S2 limited seizure, S3 full invasion.

What it found, as written at this version

  1. Under economic coercion the allies are worth nothing to Canada. The coalition model gave +8 points; the member model gives none. On average one ally sanctions or denies basing, and EU trade measures arrive in 11% of runs. The report's reason: the EU has no instrument to sanction the US on a non-member's behalf. This is the biggest correction in this version: with every module on, the chance the US gives up under coercion falls from about 40% to about 31%.
  2. In a shooting war the two models agree (+4.8 vs +4.1 in full annexation; +6.8 vs +8.0 in Alberta). Most of the value is basing denial and sanctions, not weapons: about 15 allies sanction or deny basing, 4–5 send materiel, and under one deploys.
  3. A tripwire is the strongest lever Canada and its allies control. 5,000 allied troops in Canada before a limited seizure add 7.5 points; three months' warning with token contingents adds 2. Killing allied soldiers moves the core bloc from words to basing denial.

Runs: 3,000 paired runs per variant, every other module on. 21 tests pass; all harness checks pass.

Sources are listed parameter by parameter under NATO and allies.

Full results document

0.1.2.10
12 Nov 2026
tag v0.1.2.10, commit 99f2e36
24 files, +2,633 −24 lines
“v0.1.2.10: test harness (common random numbers, paired effects, factorial screening, parameter sensitivity, in…”
Tooling

Test harness

Every variant now shares the same dice, so differences are paired and come with intervals. Automatic checks that a module is neutral when off, isolated from paths it should not touch, and pushes in the stated direction.

What it found, as written at this version

  1. The "four switches carry it" finding holds, now with tight intervals.
  2. The 2-point "effects" of wildlife and seasonal insurgency in economic coercion were noise. They are now exactly zero, run by run, which the isolation check enforces.
  3. Some effects I called noise are real but small: wildlife in the Alberta detachment (+1.5; bear watch and insects wear down a thin occupier), weather in Alberta (+0.5), logistics in full annexation (−0.5; it favours the US).

Full results document

0.1.2.9
11 Nov 2026
tag v0.1.2.9, commit bb9175e
5 files, +1,501 −1 lines
“v0.1.2.9: all-switches run (all off, all on, remove-one, add-one, max levers), switch inventory”
Analysis

All switches, measured together

Every switch measured together. All on, all off, each one removed, each one added alone.

What it found, as written at this version

  1. Together the switches move the war a lot. With everything on, the chance the US quits rises by 9 points in full annexation, 22 in the Alberta detachment and 29 in economic coercion. The US win rate in coercion is nearly halved (20% → 11%).
  2. Four switches carry almost all of it: allies, Canadian mobilization, potash and oil.
  3. Allies and mobilization matter in a shooting war.

Runs: seed 20261105, 2,000 per variant for the remove-one and add-one sets; the four headline variants were rerun at 8,000 (seed 777) to cut noise. January start. Backtests pass; 18 tests pass.

Full results document

0.1.2.8.1
11 Nov 2026
tag v0.1.2.8.1, commit 6a5eab2
“v0.1.2.8 NATO and allies (time-varying coalition response, tripwire, basing loss, NORAD break); v0.1.2.8.1 for…”
Module

Foreign small arms shipments

Arms and ammunition for the resistance from domestic stock, the black market and state suppliers. Aggregate flows only.

Switch: arms.enabled (off by default, neutral when off). Options: arms.black_market, arms.state_suppliers (auto follows the allied response reaching materiel; or force on/off), arms.us_blockade.

What it found, as written at this version

  1. Arms shipments change the body count, not the outcome. Cut off entirely, the resistance kills about 29% fewer US troops; with open state shipments it holds at full strength. The chance the US quits stays within the run-to-run noise (about ±2 points) in every variant.
  2. The clock matters more than the flow. Domestic stocks last about a year, and the median war lasts 17–20 months. Ammunition only starts to bind in the back half of a long occupation.
  3. The black market is a real floor. On its own it recovers about half the gap between isolation and full supply, and the US can't blockade its own border trade.

Runs: 2,000 per variant, seed 20261105, every earlier module on including NATO (response drawn). 18 tests pass.

Sources cited at this version (1)
  • RCMP, 2025 and 2024 Commissioner of Firearms Reports; New York Times (Dec 2025) on US-origin crime guns in Ontario and Toronto; RCMP national tracing data via PolySeSouvient (June 2026); CBC (2025).

Full results document

0.1.2.8
11 Nov 2026
tag v0.1.2.8, commit 6a5eab2
15 files, +1,342 −4 lines
“v0.1.2.8 NATO and allies (time-varying coalition response, tripwire, basing loss, NORAD break); v0.1.2.8.1 for…”
Module

NATO and allies

The allied response as a coalition of the willing. A ceiling (words, sanctions, materiel or forces) that arrives over months.

Switch: nato.enabled (off by default). When on, the module owns the external-support level and makes it change over time; when off, the scenario's fixed external.level is used as before. Options: nato.response (draw, or force diplomatic / sanctions / materiel / forces), nato.tripwire_troops, nato.us_threatens_europe.

What it found, as written at this version

  1. The allied response is worth 5–9 points to Canada. That is more than any economic lever except the oil embargo.
  2. Sanctions do most of the work. Moving from paralysis to sanctions adds 3–16 points; moving on to materiel adds less. In economic coercion, sanctions alone nearly match materiel. Allies don't need to ship weapons to matter.
  3. Allied combat forces would be decisive but are unlikely. The US quits in 79–84% of runs when they arrive, but the model gives them a 3–4% chance.

Runs: 2,000 per variant, seed 20261105, every earlier module on (no embargoes). Backtests pass; 18 tests pass.

Sources cited at this version (1)
  • NATO, "Collective defence and Article 5"; Belfer Center Article 5 explainer; AP (15 Jan 2026), Euromaidan Press (15 and 20 Jan 2026) and The Aviationist (16 Jan 2026) on Greenland; AP (May 2026) and Air & Space Forces Magazine (June 2026) on US forces in Europe.

Full results document

0.1.2.7
11 Nov 2026
tag v0.1.2.7, commit 598ed3b
10 files, +823 −4 lines
“v0.1.2.7: supply chains and logistics (US sustainment, CAF munitions without US resupply, civilian food/drug/J…”
Module

Supply chains and logistics

Sustainment for both armies and civilian supply. Interdiction of east-west corridors, munitions stocks. Aggregate only.

Switch: logistics.enabled (off by default, neutral when off), with sub-switches us_sustainment, caf_sustainment and civilian (all on by default).

What it found, as written at this version

  1. Logistics hurts both sides and decides neither. With everything on, the chance the US quits moves by at most 1–3 points in any scenario.
  2. Canada runs out of munitions early, but the conventional phase is already lost. Canada fights ~78% of the campaign short of key munitions. That makes the opening slightly cheaper for the US, but the conventional result doesn't depend on it; the war is decided by the occupation.
  3. Sabotaging supply lines cuts both ways. Interdiction holds occupier strength down ~6% on average, but the same rail lines feed Canadians. Canada pays ~0.25% of GDP for each month of full interdiction, on top of everything else.

Runs: 2,000 per variant, seed 20261105. Complex weather (January start), wildlife, potash, oil and electricity on, no embargoes. Backtests pass; 16 tests pass.

Sources cited at this version (1)
  • Auditor General of Canada, Report 24 (Oct 2026), Ammunition for the Canadian Armed Forces; CTV and CBC on the CDS's ammunition warnings; CBC (2025) and 19FortyFive (2025) on F-35 sustainment; Agriculture and Agri-Food Canada field vegetable overview 2025; UBC Canada Food Flows via CBC; federal food blueprint via The Packer (2026); Statistics Canada via CTV (2025) on pharmaceuticals; The Conversation, Fortune and CBC (Feb 2022) on the Ambassador Bridge; Transport Canada (2020), CBC, BNN Bloomberg and Energy Now on the 2020 rail blockades.

Full results document

0.1.2.6.1
11 Nov 2026
tag v0.1.2.6.1, commit e78d479
10 files, +799 −3 lines
“v0.1.2.6.1: bears and wildlife module (collisions, bears, polar-bear watch, insects, iconic incident)”
Module

Bears and wildlife

Wildlife-vehicle collisions, bear encounters and insects as a drain on an occupier. Polar bears as a story.

Switch: wildlife.enabled (off by default, neutral when off); wildlife.north_troop_share sets the occupier presence in polar-bear range. Uses the complex weather model's regions and calendar.

What it found, as written at this version

  1. Moose matter more than bears. A full occupation produces about a thousand large-animal collisions and dozens of injuries, against roughly 17 bear casualties. Both are tiny next to combat: wildlife is 0.2–0.5% of US deaths.
  2. Wildlife costs strength more than it kills. Bear watch in the north and summer insects in the boreal belt take a few percent off occupier effectiveness. That pressure is small but constant.
  3. Polar bears matter as a story, not as casualties. In the Arctic seizure, where combat deaths are ~10, a couple of incidents of US troops killing a polar bear move the chance the US quits from 28% to 31–34%. That swing comes entirely from an uncertain narrative parameter. Like blackout blame, it belongs in the narrative module (0.1.21b), not here.

Runs: 2,000 per variant, seed 20261105, complex weather on. Backtests pass; 15 tests pass.

Sources cited at this version (1)
  • Transport Canada via Canadian Underwriter (2011); BC Wildlife Collision Prevention Program; Traffic Injury Research Foundation via eDriving (2024); Samsara (April 2026); Wilder et al. (2017), Wildlife Society Bulletin 41(3):537–547 (USGS); secondary bear-attack compilations (low confidence).

Full results document

0.1.2.6
11 Nov 2026
tag v0.1.2.6, commit f3db6ae
13 files, +1,117 −12 lines
“v0.1.2.6: complex daily weather (regions, ENSO, warming, events, daylight, cold proficiency), hooks into kinet…”
Module

Complex weather

Daily weather in five regions with persistence, weighted by where the fighting is, and the cold-weather proficiency of the US force.

Switch: weather.model: simple | complex (with weather.enabled), so off / simple / complex can be compared directly. Optional weather.objective_weights per scenario.

What it found, as written at this version

  1. The simple model understated the cold by about 70%. Weighting the weather by where the fighting is (Kevin's point about northern objectives) and using realistic US cold proficiency raises US cold casualties in a January invasion from ~1,460 to ~2,530. They would be ~2,900 in a La Niña winter.
  2. Training is the US lever, not timing. Bringing the whole force to roughly 11th Airborne standard cuts cold casualties by ~60% and the campaign by six days. In practice the US can fix this only for a small part of the force.
  3. Weather still doesn't decide the war. The chance the US quits stays at 61–65% across every variant. Cold casualties are mostly non-fatal, and US opinion tracks cumulative deaths. Weather shapes the campaign's cost and length, not whether the US stays.

Runs: 2,000 per variant, seed 20261105; all other modules on, no embargoes. Backtests unchanged (Iraq invasion and occupation pass). 14 tests pass.

Sources cited at this version (1)
  • ECCC Canadian Climate Normals 1991–2020 (station weatherboxes, incl. Fort McMurray and Yellowknife); Canada's Changing Climate Report 2026, ch. 2; Environment Canada on ENSO via Western Producer (2022, 2024); CIFFC data via The Weather Network and This Magazine (2025); EIA and NEB on the 2016 Fort McMurray fire; freezing-rain climatology and the 1998 ice storm (ECCC, Canadian Encyclopedia); USU and Bloomberg (2026) on the 11th Airborne Division; Task & Purpose (2025) on Army cold-weather doctrine.

Full results document

0.1.2.5
11 Nov 2026
tag v0.1.2.5, commit 8ecd55c
11 files, +969 −4 lines
“v0.1.2.5: electricity trade module (switchable), cut options both ways, blackout attribution sensitivity; fix …”
Module

Electricity exports

Cross-border power by region. Cuts by either side, blackouts, and who gets blamed for them.

Switches: electricity.enabled (off by default, neutral when off), electricity.canada_cut (Canada stops exports), electricity.us_cut (US stops exports to Canada), electricity.provinces_in_objective.

What it found, as written at this version

  1. Cutting power does nothing reliable for Canada on its own. The price effect is tiny: about 0.5 points of US support at most, because Canadian power is under 1% of US use. What matters is the blackouts.
  2. The outcome depends on who gets blamed for the blackout. If Americans blame their president, the chance the US gives up rises from 33% to 48%. If they blame Canada, it falls to 23%. This belongs to the narrative layer (0.1.21b), not the grid, and it is the same pattern as raids on US soil: harm to American civilians tends to rally the US. The Ontario surcharge drew escalation.
  3. It is a summer weapon, not a winter one. This corrects the roadmap's earlier claim that winter power exports were Canada's biggest lever. Quebec and Ontario need their own power in winter, and Hydro-Québec cut NECEC in the January 2026 cold. The lever is biggest during New York and New England summer heat waves.

Runs: 2,000 per variant, seed 20261105; weather, seasonal insurgency, DMP mobilization, potash and oil (no embargoes) on. Backtests unchanged (Iraq invasion 160 vs 139; occupation 3,539 vs 3,500). 13 tests pass.

Sources cited at this version (1)
  • Canada Energy Regulator, 2025 Canada–US energy trade snapshot (May 2026) and 2024 snapshot; EIA Today in Energy (20 July and 5 Aug 2026); S&P Global (13 May 2026); ISO-NE market monitor via Acadia Center (Oct 2026); ISO-NE 2025 net energy for load; Hydro-Québec annual report 2024 via Energi Media; Blakes, CBC, Utility Dive on Ontario's March 2025 surcharge; BC Hydro import reporting (Electricity Forum, Global News).

Full results document

0.1.2.4
10 Nov 2026
tag v0.1.2.4, commit b942724
10 files, +859 −2 lines
“v0.1.2.4: oil module (switchable), Canadian embargo and US eastern cut options, oil excursions”
Module

Oil

Crude flows by who holds the basin, the world price, a strategic reserve release, gasoline prices and their effect on US opinion. Optional Canadian embargo.

Switches: oil.enabled (off by default; when off the engine uses each scenario's old flat economics.oil_lost_mbd), oil.embargo (Canada stops crude to the US), oil.us_cuts_east (the US stops Line 5 transit and its crude exports to eastern Canada), oil.basin_in_objective, oil.basin_share.

What it found, as written at this version

  1. Oil is Canada's biggest economic lever, but only when the US hasn't taken Alberta. In coercion an oil embargo raises the chance the US gives up from 24% to 37%, and to 42% stacked with potash.
  2. The old flat oil number was too pessimistic for the US in war. With an SPR release and markets adjusting, the US's peak GDP loss in full annexation falls from 1.8% to 1.5%, yet US withdrawal still rises, because gasoline prices hit opinion directly.
  3. Invading Alberta defuses Canada's two biggest economic levers, oil and potash. In Alberta detachment, oil adds only about 1–2 points. The model now gives the US a clear reason to move on the Prairies first.

Runs: 2,000 per variant, seed 20261105; weather, seasonal insurgency, DMP mobilization and potash (no embargo) on. Backtests unchanged (Iraq invasion 156 vs 139; occupation 3,546 vs 3,500). 12 tests pass.

Sources cited at this version (1)
  • Canada Energy Regulator, 2025 Canada–US energy trade snapshot (May 2026) and 2025 production snapshot (June 2026); EIA Today in Energy (29 July 2026); Federal Reserve Bank of Kansas City (2025); S&P Global Commodity Insights (31 Jan 2025); Trans Mountain fact sheet (July 2026); RBN Energy (Aug 2026); CAPP, Canadian imports of US crude, natural gas and refined products (Jan 2025); Fraser Institute and Canadian government filings on Line 5; Harbridge, Krosnick & Wooldridge (2016); Gupta, Pierdzioch & Tiwari (2025).

Full results document

0.1.2.3
10 Nov 2026
tag v0.1.2.3, commit ce50936
24 files, +3,108 −9 lines
“v0.1.2.3: potash module (switchable), embargo option, potash excursions”
Module

Potash

Canada's potash exports as a lever. A US farm shock, substitution from other suppliers over 6 to 18 months, an optional embargo, and whether the US holds the basin.

Switch: potash.enabled (off by default, neutral when off), potash.embargo, potash.basin_in_objective.

What it found, as written at this version

  1. Potash matters most where there is no shooting. In economic coercion it nearly doubles the chance the US gives up (13% → 24%, 28% with an embargo). With no US deaths, a slow-burn farm cost is one of the few things that moves US opinion.
  2. An embargo is cheap for Canada and costly for the US. It costs Canada about US$600M a year net of diversion and adds 4–5 points to US withdrawal in every war scenario.
  3. Invading Saskatchewan partly defuses the weapon. In full annexation, holding the basin restores some supply, so potash adds only 3–5 points to withdrawal. This gives the US an incentive to take the Prairies early — a reason the Prairies matter in the occupation plan.

Runs: 2,000 per variant, seed 20261105, weather, seasonal insurgency and DMP mobilization on. Backtests unchanged (Iraq invasion 151 vs 139; occupation 3,551 vs 3,500).

Sources cited at this version (1)
  • USGS Mineral Commodity Summaries 2026 (Potash); Natural Resources Canada, Potash facts; USDA ERS ERR-354 (2023); World Bank data via Union of Concerned Scientists (2022); Railway Age (2026); Fertilizer Daily (May and Sept 2026).

Full results document

0.1.2.2
5 Nov 2026
tag v0.1.2.2, commit af36e48
10 files, +121 −6 lines
“v0.1.2.2: switchable seasonal insurgency (fighting-season tempo, troop pressure, mud seasons); version stamp o…”
Module

Seasonal insurgency

A fighting season for the resistance (tempo, pressure on occupier troops, mud seasons). Every run now carries the model version that produced it.

No results document was written at this tag. The module's measured effect is in v0.1.2.9 and v0.1.2.10.

Sources are listed parameter by parameter under Seasonal insurgency.

0.1.2
5 Nov 2026
tag v0.1.2, commit 00e4a8a
10 files, +269 −5 lines
“v0.1.2: switchable weather and season (ECCC 1991-2020 normals, per-run anomaly)”
Module

Weather and season

Monthly weather from Environment and Climate Change Canada's 1991 to 2020 normals, with a per-run anomaly. Replaced by the daily, regional model in v0.1.2.6.

No results document was written at this tag. See v0.1.2.6 for the weather results.

Sources are listed parameter by parameter under Weather.

0.1.1
5 Nov 2026
tag v0.1.1, commit 2ca8c60
15 files, +875 −23 lines
“Add governor-campaign shock excursion”
Module

Citizen mobilization

Canada's Defence Mobilization Plan and full US mobilization with a draft, each switchable. The opening fight now handles several pools of troops of different quality.

No results document was written at this tag. The module's measured effect is in v0.1.2.9, v0.1.3.6 and v0.1.3.7.

Sources are listed parameter by parameter under Citizen mobilization.

0.1
5 Nov 2026
tag v0.1, commit 1fb740f
“Correct strategy trade-off figures in v0.1 results”
Base game

The base game: four layers

Campaign-level Monte Carlo with four coupled layers. Four Canada–US scenarios, 36,000 runs. Two backtests against Iraq, one of them in-sample.

What it found, as written at this version

  1. The conventional war is short and does not decide the outcome. It takes about 4 weeks to reach the national objectives. Canada's organized defence breaks well before that. This is robust: it holds at any plausible force ratio.
  2. The troop arithmetic doesn't close. Holding all of Canada at Quinlivan densities needs a median of roughly 490k–710k troops, against a sustainable US ceiling of about 370k without a draft. Control stays at 50–70% and the resistance feeds on the gap. This is robust: it is basic arithmetic on population and force structure.
  3. US public support at the outset is the single biggest driver. With no casus belli (polling at 2% support for force, rising to around 18% with a rally), the modal outcome is US withdrawal within about 15 months (58–67% of runs). With a manufactured crisis and a rally, the war lasts about 3 years, US deaths roughly double, and the modal outcome becomes Canadian capitulation (63–75%). Ranking runs by initial support moves the withdrawal rate from 64% in the bottom quarter to 8% in the top quarter. Armed-resistance propensity barely moves outcomes; it mostly moves casualties. The direction of this finding is robust. The capitulation share rests on the least-calibrated parameter in the model.

Sources are listed parameter by parameter under Conventional fight, Occupation and resistance, Politics in both capitals, Economy.

Full results document